Email still returns more per pound than any other channel. AI makes it faster to run well, if you keep your hand on the wheel.
Email marketing still delivers the highest return of any channel, roughly 36 pounds back for every one spent. AI does not change that maths, but it removes the two biggest bottlenecks: the blank page and the time it takes to personalise. Use it to draft, to spin up twenty subject-line variants in a minute, to write segment-specific versions, and to interpret your reports. Then edit every send by hand, because AI-written email that goes out untouched is exactly the beige, forgettable stuff people delete.
Let me put my cards on the table after a decade of running campaigns: email is still the channel I trust most. Nobody gets excited about it. It never trends. It just quietly out-earns nearly everything else, at around 36 pounds back for every pound spent by Litmus’s numbers, which is higher than any other channel.[1] So when a client tells me they want revenue rather than vanity metrics, I start with the email list every time.
So why bring AI into something that already works? Because email has always had two bottlenecks, and AI happens to be brilliant at both. The first is the blank page, staring at an empty draft on a Friday when you owe three campaigns by Monday. The second is personalisation, which everyone knows works and almost nobody does properly because writing five versions of the same email by hand is soul-destroying.
AI clears both. It gets you from nothing to a solid draft in seconds, and it will happily write eight variations for eight segments without complaining once. That does not make the strategy for you, and it will not save a bad offer. But it takes the parts of email marketing that used to eat your Friday and compresses them into minutes, which frees you up for the parts that actually move numbers.
The single biggest mistake I see is people opening ChatGPT and typing write me a marketing email. You get generic sludge because you gave it nothing to work with. Strategy comes first, and AI is a genuinely useful thinking partner for it, as long as you lead.
Before you write a word of copy, get clear on three things: who this email is for, what one action you want them to take, and where they are in their relationship with you. A welcome email to a brand-new subscriber and a win-back to someone who has ignored you for six months are completely different jobs. Feed that context in. Try: I run email for a [type of business]. I want to plan a five-email welcome sequence for new subscribers who just downloaded [lead magnet]. Map out what each email should accomplish and the one call to action for each. Now you are planning, not just prompting.
This is also where AI helps you avoid the classic trap of blasting everyone the same thing. Ask it to help you think through your segments and what each one needs to hear. If you want to go deeper on the strategic side, how to use AI for marketing covers the wider picture, and the sequence thinking here sits right alongside AI newsletter marketing.
Never ask AI to write an email until you can tell it who it is for and what one thing you want them to do. Context is the difference between generic and effective.
Most AI-written email is instantly recognisable, and not in a good way. It reads smooth and a bit too keen, every sentence the same tidy length, and then it slides straight out of your memory. If your open rates are fine but nobody ever hits reply, this is usually why. The fix sits in two places: how you prompt, and how you edit afterwards.
Prompt for a voice, not just a message. Compare write an email about our new feature with write a short, warm email to existing customers about our new scheduling feature, in a friendly, plain-spoken voice, no hype, no exclamation marks, like a helpful colleague sending a quick heads-up. The second gives AI the constraints it needs to sound human. Feed it examples of emails you have sent that landed well, and ask it to match that tone. Better still, train it on your actual brand voice so it stops defaulting to bland; our guide on writing a newsletter with AI without sounding like a robot goes deep on this.
Then edit like you mean it. I never send an AI draft untouched, and neither should you. Cut the throat-clearing opening line, which is almost always deletable. Swap one generic phrase for something specific to your audience. Read it aloud, and anywhere it sounds like a press release, rewrite it the way you would actually say it. That final human pass takes about three minutes. It is also the whole reason your email sounds like you rather than like everyone else’s untouched AI draft. If you want ready-made starting points, ChatGPT prompts for email marketing gives you a set to adapt.
If you use AI for one thing in email, make it subject lines. This is the task it is almost perfectly suited to, because subject lines reward volume and variety, and generating variety is exactly what AI does effortlessly.
Writing subject lines by hand, you might grind out four before you run dry and pick the least bad one. Ask AI for twenty, across different angles, and you get options you would never have reached alone: give me twenty subject lines for this email, mix curiosity, urgency, benefit-led, and plain-direct styles, keep them under fifty characters, no clickbait. In under a minute you have a menu.
Then test properly, because your instinct about subject lines is worse than you think, and so is mine. Most email platforms have built-in A/B testing. Pick your two or three favourites from the AI list, split-test them on a slice of your list, and send the winner to the rest. Over a few campaigns you learn what your specific audience responds to, which is worth more than any best-practice list. We broke this exact loop down in how to A/B test 20 email subject lines in 10 minutes with AI.
Generate twenty subject lines with AI, shortlist three, and let a real A/B test pick the winner. Volume from AI plus testing from your platform beats guessing every time.
Personalisation is the thing everyone agrees works and most teams still fake with a merge tag in the greeting line. Real personalisation means the content itself changes for different people, and this is where AI quietly changes the economics.
The reason teams do not do it is effort. Writing a genuinely different version of a campaign for your enterprise buyers, your small-business buyers, and your lapsed customers used to mean writing three emails. Now you write one strong version, then prompt: rewrite this same email for [segment], keeping the offer identical but changing the framing, examples, and pain points to fit them. Three tailored versions in the time it used to take to write one.
Start with the segments that actually matter rather than slicing your list into confetti. Usually that means splitting by where someone is in the journey (new, active, lapsed) or by an obvious difference in what they need. Even two or three well-chosen segments, each getting a version that actually speaks to them, will beat one email to everyone. The data backs the effort: campaigns that weave AI-driven personalisation through the workflow consistently report meaningfully higher revenue per recipient than batch-and-blast sends.[2] Just keep a human reading each version, because a mistargeted personalised email lands worse than a generic one.
Most marketers look at their email report, note that the open rate was fine, and move on. That is a wasted asset. Your analytics are a map of what your audience wants, and AI is good at reading maps if you hand it the data.
Export or copy your campaign stats and paste them in: here are the open rates, click rates, and subjects from my last ten campaigns. What patterns do you see, and what should I test next? It will spot things you glazed over, that your how-to subject lines beat your promotional ones, or that Tuesday sends outperform Thursday. Those are hypotheses, not laws, but they point your next test in a smarter direction.
You can do the same with the content itself. Ask it which of your recent emails likely drove the most engagement and why, then double down on that format. The point is to close the loop: every campaign teaches you something, and AI helps you actually extract the lesson instead of letting it evaporate. Do not outsource the judgement, though. AI can spot a pattern in the numbers, but you know whether that spike was your subject line or the fact that you ran a sale that week.
One more use worth building in: send-time and cadence. Paste in when you have been sending and how each send performed, and ask AI to suggest a testing plan for timing and frequency rather than guessing. It will not know your audience better than you do, but it is good at turning a vague hunch like maybe we email too often into a structured experiment you can actually run and measure. That turns your reporting from a rear-view mirror into a plan for the next quarter.
I will finish with the tells, because avoiding them is most of the battle. When an email screams AI wrote this, it is almost always one of these.
None of this means avoid AI. It means use it as the fast first-drafter and the tireless variant-generator it is, then bring the human back for the edit that makes it land. That balance, machine speed with human judgement, is the whole game in email right now. If you want your whole team working this way, our corporate AI training is built around practical, workflow-first marketing use exactly like this.
Use it across the workflow but keep a human in charge. Plan the strategy and segments with it, draft your copy, generate twenty subject-line variations to A/B test, write tailored versions for each segment, and paste in your analytics to spot what to test next. The one firm rule is that AI drafts and a human always edits before sending, because untouched AI email is generic and readers can tell.
Only if you send them untouched. Raw AI copy tends to be polished but generic, over-enthusiastic, and slow to get to the point, which suppresses replies. When you prompt it for a specific voice, feed it examples of your best emails, and do a three-minute human edit on every send, AI-assisted emails perform as well as or better than fully hand-written ones, at a fraction of the time.
Subject lines. It is the task AI is almost perfectly suited to, because good subject-line work rewards generating lots of varied options and then testing them. Ask for twenty variants across different angles, shortlist your favourite two or three, and run a real A/B test in your email platform. That loop reliably beats hand-writing four subject lines and guessing which is best.
Yes, and this is where it changes the economics of email. Write one strong version, then ask AI to rewrite it for each segment, keeping the offer the same but changing the framing, examples, and pain points. That gives you three or four genuinely tailored versions in the time it used to take to write one, which is why personalised, segmented campaigns consistently out-earn one-size-fits-all sends.
Very much so. Email still returns roughly 36 pounds for every pound spent, higher than any other channel, and you own the list rather than renting an audience from a platform. AI does not replace that advantage, it strengthens it by removing the two biggest time costs, the blank page and personalisation, so a small team can run the kind of tailored, well-tested email programme that used to need a much bigger one.
This guide draws on more than a decade of hands-on email campaign work, alongside current benchmarks from Litmus on email marketing ROI and industry data on AI-driven personalisation performance. All figures are sourced and linked below.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.